Acoustic characteristics of tense and lax vowels across sentence position in clear speech Open Access

Clear speech is a speaking style people adopt when they are aware of a possible difficulty in the listener's speech perception due to, for example, a hearing deficit, background noise, language level, or a different native language (Smiljanic and Bradlow, 2005). A number of studies have investigated effects of clear speech on improving speech intelligibility for various listener populations, including hearing-impaired and normal-hearing adult listeners (Picheny et al., 1985), non-native adult listeners (Bradlow and Bent, 2002), and school-aged children with and without learning disabilities (Bradlow et al., 2000). The current study aims to extend our understanding of clear speech by providing an investigation of the acoustic characteristics of tense and lax vowels across sentence position in clear and conversational speech. This investigation is fundamental, especially because there has been controversy surrounding the definitions of tense and lax vowels (e.g., Stevens, 2000). Clear speech provides a unique environment in which to study the inherent properties of tense and lax vowels, because it provides a context where speakers make exaggerated articulations to produce vowels (through lengthening and vowel space expansion) (Ferguson and Kewley-Port, 2007; Ferguson and Quené, 2014; Picheny et al., 1986). The known characteristics of clear speech allow us to easily predict that tense vowels will be more lengthened and peripheral than usual in clear speech, because such changes will maximize their inherent properties. In contrast, lax vowels are in general characterized by shorter duration and centralization, the characteristics that run contrary to clear speech characteristics. Therefore, the question arises as to how the inherent properties of lax vowels change in clear speech in order for them to be more intelligible in this context. Will lax vowels be lengthened and become peripheral in clear speech just like tense vowels? If so, will the amount of lengthening and decentralization be comparable between tense and lax vowels?

So far, the majority of studies on clear speech have focused on vowel group trends rather than vowel specific trends, and only a handful of studies examined how lax vowels change in clear speech as compared to tense vowels. Existing studies that examined tense lax differences also have presented conflicting findings, perhaps due to differences in stimuli type and phonetic context. With respect to temporal measurements, some studies indicated that tense and lax vowels were lengthened to a similar degree from conversational to clear speech (Smiljanic and Bradlow, 2008), whereas others indicated that lax vowels increased much less in duration than tense vowels (Leung et al., 2016; Picheny et al., 1986). With respect to spectral measurements, some studies showed that the vowel space areas were similarly expanded in clear speech for both tense and lax vowels (Leung et al., 2016), while others indicated that lax vowels were expanded more than tense vowels (Picheny et al., 1986). Besides vowel space areas derived from the formant values extracted from one time point, some studies examined spectral changes (which involved changes of formant values over time) of tense and lax vowels. These studies revealed that lax vowels showed a greater spectral change in clear speech compared to tense vowels (Lam et al., 2012; Leung et al., 2016). In particular, Leung et al. (2016) examined the contrast between three pairs of tense and lax vowels (/i-ɪ/, /ɑ-ʌ/, /u-ʊ/) which appeared in isolated /kVd/ words as in kid. In their study, tense vowels showed a greater durational increase in clear speech, whereas lax vowels exhibited greater spectral changes. Therefore, tense and lax vowels differed in the type of modification they underwent in clear speech.

Current research on vowel production in clear speech also lacks any systematic manipulation of the sentence position in which target vowels appear. It has been widely documented that the position of the target word in its sentence affects its acoustic characteristics (Oller, 1973). In particular, words at the end of prosodic domains are found to be lengthened, a phenomenon known as final lengthening. Thus, sentence-final position could provide an additional context for speakers to emphasize vowels. Smiljanic and Bradlow (2008) suggested that speech style interacted with sentence positions. More specifically, they found that the durational contrast between vowels preceding voiced vs voiceless codas increased in clear speech in sentence-final position only. At the same time, vowel duration difference between clear and conversational speech was significant only in sentence-medial position. The present study builds on previous work by examining clear speech effects on the production of tense vs lax vowels and its interaction with sentential position.

Speech recordings were collected from 12 participants (6 males and 6 females, 18–28 yr of age). All of them were monolingual speakers of American English, and reported that they grew up in the Midwest. None of them had a history of a speech, language, or hearing problem. During the experiment, participants watched a video where they were introduced to two “listeners” who gave instructions asking them to read sentences appearing one at a time on a screen. Of the women who acted as “listeners,” one was an elderly woman and the other was a young college student. The elderly “listener” asked the participants to read the sentences as clearly as possible because she had a hearing loss. The young “listener” asked the participants to read the words casually or as they normally would because she had no problem with hearing. It was expected that the instructions from the elderly “listener” would prompt speakers to shift into the use of clear speech, whereas the instructions from the young “listener” would prompt conversational speech. The order of the presentation of the two “listeners” was randomized so that half of the subjects received the clear speech condition first, and the other half received the conversational speech condition first.¹ Within each condition, the order of the sentences was also randomized.

The tense and lax vowels examined in this study included the monophthongal vowels /i, u, ɑ/ and /ɪ, ɛ, æ, ʊ, ʌ/, respectively. For each of the 8 vowels, there were 4 words containing the target vowel with a total of 32 words. When choosing the words, we also considered their word frequency and neighborhood density ratings by referring to English Lexicon Project database (Balota et al., 2007). The target words were embedded in 32 sentences containing two target words each (one in the medial position and one in the final position) as in: The sick boy received a shot (target words are underlined). All stimuli were simple sentences containing one independent clause, having a similar length between six and eight syllables. In addition, there were eight filler sentences.² The final data set included 3067 word tokens (12 participants × 32 sentences × 2 target words × 4 repetitions - 5 excluded tokens).

To determine the amount of vowel change in clear speech, we used vowel duration and three measures of vowel hyperarticulation: vowel space area, vowel space dispersion, and vowel peripheralization (following Smiljanic and Bradlow, 2005). The vowel space area was measured as the Euclidean area covered by the triangle for the tense vowels /i, u, ɑ/ or the pentagon for the lax vowels /ɪ, ɛ, æ, ʊ, ʌ/ made from the average F1 and F2 (measured at the vowel midpoint) coordinates of each vowel category. Vowel space dispersion was calculated as the average distance of each vowel from the geometric center point (i.e., centroid) in the speaker's F1 by F2 space. When a talker produces a vowel in a clearer manner, the resulting vowel will subsequently move further away from the center point. Vowel peripheralization in clear speech relative to conversational speech for each vowel category was calculated as the Euclidian distance between two points (i.e., an average F1, F2 coordinate point in conversational speech and an average F1, F2 coordinate point in clear speech) in the F1 by F2 plane. This distance represents the amount of change an individual vowel underwent from its conversational speech form to its clear speech form.

We used a 3-way repeated measures analysis of variance (ANOVA) to examine the effects of style (conversational vs clear), tenseness (tense vs lax), and position (sentence-medial vs sentence-final) on vowel duration. As expected, the main effect of style was significant, with longer vowels in clear speech than in conversational speech (see the Vowel duration column of Table 1 for statistics for vowel duration). Although the main effect of position was not significant, the interaction effect of style × position was significant. Pairwise t-tests with Bonferroni correction (adjusted α = 0.05/4 = 0.0125) revealed that the vowel duration of clear and conversational speech differed in sentence-medial position (p < 0.001), but not in sentence-final position (p = 0.02) (see Fig. 1). Furthermore, final vowels showed a tendency to be longer than medial vowels in conversational speech, although the difference did not reach statistical significance after the Bonferroni correction (p = 0.03). Final lengthening on vowels was not observed in clear speech (p = 0.22).

The main effect of tenseness was significant, with longer duration for tense vowels (M = 142 ms) than for lax vowels (M = 135 ms). The effect of tenseness held true for both clear and conversational speech, as shown by the absence of a significant interaction between tenseness and style. Furthermore, the tense-lax vowel distinction was equally present in the medial position and the final position, as shown by no significant interaction between tenseness and position. Finally, three-way interaction of style × tenseness × position was not found to be significant.

While lax vowels are generally shorter than tense vowels, the lax vowel /æ/ is known to be one of the longest vowels in American English. This was also true in our data, as shown by the following average duration values for each of the vowels across speech style and sentence position. Durations (in ms) are shown in the parentheses, from the longest to the shortest: /æ/ (189), /ɑ/ (163), /ɛ/ (148), /u/ (140), /ʌ/ (127), /i/ (122), /ʊ/ (105), and /ɪ/ (104). To rule out the possibility that /æ/ obscured overall lax vowel patterns, we ran additional ANOVA with /æ/ excluded. The significance of the effect of each factor did not change after excluding /æ/, suggesting that our results still held for the rest of the lax vowels.

Separate 2-way repeated measures ANOVAs were performed for tense and lax vowels to examine the effects of style and position on respective vowel space areas, which differed in their shapes. As shown in the Vowel space area column of Table 1 (T is for tense vowels and L is for lax vowels), the main effect of style was significant. As expected, the clear speech style elicited a greater vowel space area for both tense and lax vowels. The main effect of position was found to be significant only for lax vowels. As shown in Fig. 2, lax vowel space areas were greater in medial position than in final position. This was true in both conversational and clear speech, as shown by insignificant style × position interaction.

A 3-way repeated measures ANOVA was used to examine the effects of style, tenseness, and position on vowel space dispersion. As shown in the Vowel space dispersion column of Table 1, the main effect of style was significant, where vowels in clear speech had greater dispersion than those produced in conversational speech. Although the main effect of position was found to be insignificant, the interaction effect of style × position was significant. It can be seen in Fig. 3 that the amount of increase in vowel space dispersion from conversational to clear speech is greater for sentence-final vowels. This observation was confirmed by pairwise t-tests. Vowel space dispersion of clear and conversational speech differed significantly in sentence-final position (p < 0.001), but not in medial position (p = 0.03) after Bonferroni correction. In addition, the difference in vowel space dispersion between sentence-medial and sentence-final vowels was not significant in clear speech (p = 0.16), nor in conversational speech (p = 0.67).

The effect of tenseness was significant. Expectedly, tense vowels, which are produced at more extreme ends in the oral cavity, had a greater degree of vowel space dispersion than lax vowels did. Average vowel space dispersion values (in Hz) for each of the vowels across speech style and sentence position are shown in parentheses, from the largest to the smallest: /i/ (742), /ɑ/ (505), /u/ (362), /ʊ/ (362), /ɪ/ (354), /ʌ/ (213), /ɛ/ (188), and /æ/ (170). The tenseness × style interaction was not significant, suggesting that the distinction between tense and lax vowels remained stable across the speech styles. The tenseness × position and style × tenseness × position interactions were not significant either.

A 2-way repeated measures ANOVA was performed to examine the effects of tenseness and position on the extent of peripheralization in clear speech relative to conversational speech. As for the main effect of tenseness, there was no significant difference found for the amount of vowel peripheralization between tense and lax vowels, suggesting tense vowels and lax vowels peripheralized to a similar degree in clear speech (see the Peripheralization column of Table 1). Average vowel peripheralization values (in Hz) for each of the vowels across sentence position are shown in parentheses, from the largest to the smallest: /u/ (72), /æ/ (63), /ɪ/ (55), /ɑ/ (54), /ʌ/ (50), /i/ (49), /ʊ/ (49), and /ɛ/ (42). Neither position nor tenseness × position was significant.

The present study showed that both tense and lax vowels were lengthened and decentralized to a similar degree in clear speech. Smiljanic and Bradlow (2008) also indicated that tense and lax vowels were similarly lengthened, and their study and our study share the method that the target words, varying in segmental contents, appeared in sentential contexts (rather than in isolation as in Leung et al., 2016), although all of their target words appeared sentence-medially. We also showed that both tense and lax vowels' vowel space areas were expanded, and that there was no difference in the amount of vowel space dispersion between the two vowel categories, supporting Leung et al. (2016). Our study further extends the findings by showing a comparable increase in vowel peripheralization between tense and lax vowels in clear speech.

We would like to point out two main findings with respect to the effect of sentence position on vowel production. First, there was a significant effect of position on the vowel space area of lax vowels, with greater vowel space expansion in the medial position. As Fig. 2 indicates, the vowel space expansion for lax vowels in medial position appears to be primarily driven by the decreased F1 in high vowels /ɪ, ʊ/. The literature has shown that F1 generally increases across vowels in clear speech, although this effect is greater for low vowels than for high vowels (Ferguson and Kewley-Port, 2007). Increased F1 has often been attributed to a lower tongue position, which might reflect the increase in vocal effort, loudness (Ferguson and Quené, 2014), and sonority (Beckman et al., 1992; de Jong et al., Beckman, and Edwards, 1993) in clear speech. Under these accounts, the lower F1 for high lax vowels in medial position might in fact reflect an increase in reduction. Therefore, while an overall vowel space expansion is expected for clear speech, our findings suggest that vowel space expansion could interact with high lax vowels if their F1 is lowered in reduced speech. At the same time, more research is needed to better understand how the clarity of high vowels is achieved, as the findings on F1 are contradictory for clearly produced high vowels (de Jong et al., 1993; Erickson, 2002; Ferguson and Quené, 2014).

Second, the effect of position alone was not enough to elicit longer durations or greater vowel space dispersion; rather, it was the interaction between style and position that determined vowel properties. The difference in vowel duration between clear and conversational speech was greater in sentence-medial position. As shown in Fig. 1, the duration of medial vowels increased significantly from conversational to clear speech, while the duration of final vowels, which was already long in conversational speech, increased only slightly in clear speech. The finding reinforces Smiljanic and Bradlow (2008) who showed a significant clear speech effect on vowel duration only in the medial position. We found that the interaction between style and position was also significant on vowel space dispersion, but the difference between clear speech and conversational speech was greater in sentence-final position (see Fig. 3). A closer examination of the data revealed that in clear speech, vowel space dispersion tended to be greater sentence-finally compared to sentence-medially, mostly due to tense vowels. In conversational speech, however, vowel space dispersion tended to be greater sentence-medially than sentence-finally, particularly because lax vowels exhibited significantly greater vowel space dispersion in the medial position. This resulted in increased gaps between clear and conversational speech in the final position. The different patterns for tense and lax vowels between sentence positions are also reflected in a trend for an interaction between tenseness and position (p = 0.07), as shown in Table 1 (see also endnote ³). Unlike tense vowels, lax vowels were slightly more dispersed in the medial position compared to the final position. As a result, the difference in vowel space dispersion between the two vowel types increased sentence-finally than sentence-medially, although tense vowels had significantly larger vowel space dispersion across sentence positions relative to lax vowels.

Overall, our vowel duration and hyperarticulation measurements indicate that lax vowels were manipulated to a similar degree as tense vowels in clear speech. This suggests that these adjustments are part of a general process in clear speech. The speakers also used vowel-specific strategies in different sentence positions; unlike tense vowels, lax vowels showed a significantly increased vowel space area and a tendency for greater vowel space dispersion in sentence-medial position. Given that the defining properties of lax vowels (namely their shorter duration and centralization) are affected by speech style and sentential position, future studies should determine how listeners make use of these different modifications in various contexts.